When does an organisation using AI become the provider of a system?

Why the question comes up at all

Artificial intelligence rarely enters an organisation by a board decision. It arrives as a new feature in software bought years earlier, or as a tool built in-house on a general-purpose model. In neither case does anyone ask about the role, because nobody feels they are placing anything on the market or putting anything into service.

The Regulation does not ask whether an organisation is in the AI business. It asks about the role towards a specific system. The same organisation can be a deployer of one tool and a provider of another.

On top of that sits the confusion around the July 2026 change to the timetable. The postponement covered Chapter III Sections 1, 2 and 3, with the exception of Article 6(5) on the Commission’s classification guidelines. Those are classification, requirements, and the obligations of providers and deployers of high-risk systems. It did not cover the prohibited practices or the AI literacy duty. The transparency obligations in Article 50 have applied since 2 August 2026. The same amendment did, however, give providers of systems generating synthetic content that were placed on the market before 2 August 2026 until 2 December 2026 to comply with Article 50(2) (Article 111(4)). The market took one sentence from this — “the AI Act has been postponed” — and stopped counting deadlines.

What to establish before deciding

When an organisation is a provider from the outset

The definition of a provider in Article 3(3) does not require the system to be sold or made available externally. Putting into service includes supplying a system for own use (Article 3(11)). A model supplied by another company does not take the provider role away from the organisation (Article 3(68)). In its guidelines on Article 50 the European Commission gives the example of an organisation that has developed a chatbot in-house and puts it into service for its own use under its own name (paragraph 11). According to the Commission, such an organisation is a provider. The risk category does not matter for this role.

The obligation in Article 4 has bound the provider since 2 February 2025. From 2 August 2026 the provider ensures that a system intended to interact directly with people is designed and developed so that they are informed that they are interacting with an AI system (Article 50(1)). Among other exceptions, the obligation in Article 50(1) does not apply where the interaction with AI is obvious to a reasonably well-informed, observant and circumspect person. Obviousness is assessed taking into account the circumstances and the context of use. The provider of a system generating synthetic text, images, audio or video ensures that the outputs are marked in a machine-readable format (Article 50(2)). The exceptions in Article 50(2) include an assistive function for standard editing. Providers of such systems placed on the market before 2 August 2026 have until 2 December 2026 to comply with the marking obligation (Article 111(4)). The provision mentions only systems placed on the market. In its guidelines on Article 50 the European Commission describes that period as also covering systems put into service before 2 August 2026 (paragraph 153).

Three circumstances in which the role changes

Article 25(1) names the distributor, the importer, the deployer and any other third party. Each of them is considered a provider of a high-risk AI system if any of three circumstances arises.

Own name or trademark. The organisation puts its brand on a high-risk system already placed on the market or put into service. The provision attaches a qualifier to this circumstance. This applies without prejudice to contractual arrangements allocating obligations differently.

Substantial modification. The organisation makes a substantial modification to a high-risk system already placed on the market or put into service. After the modification it remains a high-risk system under Article 6. The point is therefore a change within the same category, not an exit from it.

Change of intended purpose. The organisation changes the intended purpose of a system that was not classified as high-risk and has already been placed on the market or put into service. After the change it becomes a high-risk system under Article 6. The provision names general-purpose AI systems explicitly.

The third circumstance is the hardest to notice. The HR department starts using a purchased general-purpose assistant to evaluate job candidates (Annex III, point 4(a)). The system has not changed. Its intended purpose has, and with it the role of the organisation. It is different when a team has built the assistant itself on another company’s model. The organisation is then its provider from the outset (Article 3(3)). The new purpose changes the category of the system, not the organisation’s role.

Article 25(3) contains a separate rule for the manufacturer of a product covered by the Union harmonisation legislation listed in Section A of Annex I. The manufacturer is considered the provider of a high-risk AI system that is a safety component of its product. This happens when the system is placed on the market together with the product or put into service after the product has been placed on the market, in both cases under the manufacturer’s name or trademark.

What the original provider owes the new one

Article 25(2) sets out what happens to the original provider. After the change of role the original provider is no longer considered the provider of that particular system. In exchange it has a duty of close cooperation with the new provider.

The provision names, by way of example, three elements of that cooperation where they are relevant to its purposes.

• technical documentation sufficient to assess conformity with the requirements in Article 16;
• information about the known limitations and failure modes of the system;
• targeted technical access, including for testing and validation purposes.

For an organisation whose role has changed, this is the main source of material for performing the provider obligations. Without the original provider’s documentation the conformity assessment is difficult to carry out. Article 25(5) preserves the protection of intellectual property rights, confidential business information and trade secrets.

The provider’s statement that switches off the duty of cooperation

The last subparagraph of Article 25(2) closes the loop. The duty of cooperation does not apply where the original provider has clearly specified that its system is not to be changed into a high-risk AI system. In that case it owes neither cooperation nor documentation. The provision does not require that statement to be in a contract.

The consequence is practical and unwelcome. An organisation using a system covered by such a statement takes on the whole risk of a change of purpose. After such a change it is left with provider obligations and without the material to perform them. The statement does not protect it from the change of role. What it does is remove the support.

This is why the provider’s statement should be verified at procurement rather than at deployment. Article 25(4), which the July 2026 amendment extended to AI models, points the same way. The obligation applies to the provider of a high-risk AI system and to a third party that supplies elements used in or integrated into that system. Among them the provision lists an AI system and an AI model. It also lists tools, services, components or processes. The parties specify in a written agreement the information, capabilities, technical access and other assistance based on the generally acknowledged state of the art. The point is what the provider needs to fully comply with its obligations under the Regulation. Tools, services, processes and components released publicly under a free and open-source licence are excluded. The exclusion does not cover general-purpose AI models.

When this binds, and why the date does not excuse waiting

Articles 25 and 26 sit in Chapter III Section 3, so the postponement covers them. They start to apply on 2 December 2027 for high-risk systems under Annex III and on 2 August 2028 for systems under Annex I.

For operators of high-risk AI systems placed on the market or put into service before those dates, the Regulation applies only after a significant change in the design of those systems (Article 111(2)). The prohibitions in Article 5 apply to those systems without that limitation. In any case, providers and deployers of high-risk AI systems intended to be used by public authorities must take the necessary steps to comply with the Regulation by 2 August 2030. Whether a change of intended purpose alone counts as a significant change in design is not expressly settled by the provision. For each use case it is therefore worth recording the date on which the system was first placed on the market or put into service.

The date from which Article 25 will apply is not a start date for the work. It is the date by which the work has to be finished. Contracts with AI vendors signed today will often still be in force in December 2027. They will decide whether the organisation has access to documentation at the moment the provision starts to bind.

It is worth remembering separately that part of the Regulation already applies. The AI literacy duty in Article 4 has applied since 2 February 2025, and in its current wording since 27 July 2026. The prohibitions in Article 5 have applied since 2 February 2025. Two new prohibitions apply from 2 December 2026. The transparency obligations in Article 50 have applied since 2 August 2026. Providers of systems generating synthetic content that were placed on the market before that date have until 2 December 2026 to comply with Article 50(2).

Most common mistakes

• Asking about the role once, at the level of the organisation. The role attaches to a system, not to a company. Establishing that “we are a deployer” without naming the tool settles nothing.
• Assuming that buying a ready-made tool rules out the provider role. Any of the three circumstances in Article 25(1) makes the organisation the provider of a high-risk AI system, not only a change of purpose. An organisation that builds its own tool on a purchased model and puts it into service under its own name is also a provider. Such decisions are sometimes made by a product team rather than by the board.
• Reading the postponement as a suspension of the whole Regulation. The postponement covered classification and the requirements and obligations concerning high-risk systems. The remaining provisions apply from the dates in Articles 111 and 113, and most of them already apply.
• Reading Article 25(1) without the contractual qualifier in point (a) and without the exclusion in paragraph 2. Both sentences change the outcome of the analysis, and both are easy to miss on a quick reading.
• Postponing the contract review until the provision starts to bind. By December 2027 the terms will already have been agreed, and renegotiating an agreement takes longer than concluding one.

Conclusions and next steps

Establishing the role is a repeated exercise, not a one-off. It repeats with every new use case and with every change in how an existing one is used.

An order that keeps this work manageable.

1. List the AI use cases in the organisation together with their intended purpose. Without that list the question about the role has nothing to attach to.
2. For each use case, establish whether the organisation developed the system itself or had it developed and placed it on the market or put it into service under its own name. Then check whether any of the three circumstances in Article 25(1) that make the organisation the provider of a high-risk AI system arises.
3. Check the contracts, the instructions for use and the vendor’s documentation for a statement excluding conversion into a high-risk AI system. Such a statement changes the risk assessment of the project.
4. For new contracts, agree the scope of documentation and technical access before signing. For high-risk AI systems, Article 25(4) provides for such a written agreement.
5. Build the role assessment into the procedure for launching new tools, so that a change of purpose does not happen outside the board’s knowledge.

The evidence that this work was done is a register of use cases with an assigned role and a record of the assessment behind each one. The same material answers questions from clients, insurers and auditors.

Related materials

• What should an AI use register contain? — the list that makes the question about the role possible to ask.
• Who is responsible for AI Act transparency obligations? — the second axis dividing duties between provider and deployer.
• Does every organization fall under NIS2? — the same qualification logic in the other regime.
• What new prohibitions does the AI Act introduce from 2 December 2026 and whom do they cover? — when a provider of a generative system infringes the new Article 5 prohibitions and which safeguards protect it.

Sources

• Regulation (EU) 2024/1689 (AI Act), consolidated text as of 27 July 2026, Article 3(3), (11) and (68), Articles 4, 5, 6, 16, 25 and 26, Article 50(1) and (2), Article 111(2) and (4), Article 113, Annex I, Section A and Annex III, point 4(a) — eur-lex.europa.eu.
• Regulation (EU) 2026/1744 of 8 July 2026 amending Regulation (EU) 2024/1689 — OJ L, 2026/1744, 24.7.2026.
• European Commission, guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of Regulation (EU) 2024/1689 — content approved by Communication C(2026) 5054 final of 20 July 2026, formal adoption announced once all language versions are available — paragraphs 11 and 153 — digital-strategy.ec.europa.eu.
• Act of 3 July 2026 on artificial intelligence systems (Journal of Laws 2026, item 1003) (in Polish).